I'm going to get around to testing this out tomorrow. I just breezed through the docs and I have one comment. Can we not use jet as the default colormap? I know it is the default colormap in matplotlib but we aren't bound by that. There are dozens of sources out there explaining why jet is a bad choice. I'll give two of the ones that have popped out to me over the years:

• https://jakevdp.github.io/blog/2014/10/16/how-bad-is-your-colormap/
• http://web.stanford.edu/~mwaskom/software/seaborn/tutorial/color_palettes.html

There is also a lively Github issue over here: matplotlib/matplotlib#875

I'll try to get a few more comments out tomorrow. So far, this work is really encouraging.

I agree that jet isn't a good color scheme (this scipy talk is really
good https://www.youtube.com/watch?v=xAoljeRJ3lU), but that is easily
changed by setting cmap='whatever'.

One colorbar issue I find consistently annoying is having to manually
specify vmin and vmax or levels when plotting anomaly fields on a
diverging colorbar with white in the middle. Some sort of
symmetric_colorbar= <middle point value> kwarg would be really nice.

@jhamman I'd be happy to bundle viridis with xray as the default colormap (the source code is here) -- or we could just wait a few months, at which time it will be the matplotlib default.

@nbren12 Seaborn has some heuristics for picking sequential vs diverging colormaps. Possibly worth borrowing: https://github.com/mwaskom/seaborn/blob/v0.6.0/seaborn/matrix.py#L158

My inclination is to merge this on master and iterate in future PRs. There are still a few things to fix, but this is a great start.

yeah, +1 for the viridis colormap.

@clarkfitzg - I just spent some time running through your examples and trying to plot some of my own data. Overall - this is really slick and I really liked what you have going here. Below are a few comments I made along the way - take them or leave them.

1. Consider yincrease=True as the default. Maybe there is a downside to this but I don't see it. I'm pretty sure it is just the convention of imshow.
2. It may be worth mentioning why you are using imshow over pcolormesh. The limitation of imshow having regularly spaced coordinates will be important to some (myself included). We may, at some point, want to add an option for using pcolormesh. This would support irregular grids without forcing the use of contourf.
3. More examples with cartopy would be good. Maybe we can put together example page here: http://xray.readthedocs.org/en/latest/examples.html
4. I continue to have cartopy install problems on osx, even with python2.7 using anaconda. I also had problems on a clean docker image. I know this has nothing to do with this PR but the problems with installing cartopy will limit its applicability here.

Have you tried the ioos conda channel for cartopy? They do a really nice job maintaining builds.

On Sat, Jul 18, 2015 at 11:30 AM, Joe Hamman [email protected]
wrote:

@clarkfitzg - I just spent some time running through your examples and trying to plot some of my own data. Overall - this is really slick and I really liked what you have going here. Below are a few comments I made along the way - take them or leave them.

1. Consider yincrease=True as the default. Maybe there is a downside to this but I don't see it. I'm pretty sure it is just the convention of imshow.
2. It may be worth mentioning why you are using imshow over pcolormesh. The limitation of imshow having regularly spaced coordinates will be important to some (myself included). We may, at some point, want to add an option for using pcolormesh. This would support irregular grids without forcing the use of contourf.
3. More examples with cartopy would be good. Maybe we can put together example page here: http://xray.readthedocs.org/en/latest/examples.html

1. I continue to have cartopy install problems on osx, even with python2.7 using anaconda. I also had problems on a clean docker image. I know this has nothing to do with this PR but the problems with installing cartopy will limit its applicability here.

Reply to this email directly or view it on GitHub:
#466 (comment)

Thanks for all the feedback! I'll change the default colormap and look into the other things mentioned on Monday.

Re Python 3 compatibility with xray.DataArray(d_ylog, zip(('y', 'x'), (y, x))) it would be nice if the DataArray constructor accepted an iterable. Thoughts?

Re Python 3 compatibility with xray.DataArray(d_ylog, zip(('y', 'x'), (y, x))) it would be nice if the DataArray constructor accepted an iterable. Thoughts?

Agreed. Want to add a fix?

Sure. I'll do it separate from this PR.

Let's follow up on this with some smaller PRs.

@jhamman

1. The issue with yincrease=True as the default is that actual PNG images when loaded will be 'upside down' when plotted with xray.
2. Definitely could add support for pcolormesh. Went with imshow since it seems more common.
3. +1
4. I recently installed cartopy on a new machine with conda:

conda install -c scitools cartopy

Seems like it used an old version of matplotlib, but it still worked.
Followed their docs: http://scitools.org.uk/cartopy/docs/latest/installing.html